AutoSynthData: Generating Training Data for Enterprise Agents

Synthesis

1. 01

   Difficulty-calibrated task selection

   For training, tasks should expose current agent weaknesses—i.e., be feasible and realistic but not yet consistently solved. Reliably solved tasks provide little new training signal.

   Supporting evidence 1

   Original excerpt

   Difficulty. For training, the task should expose a weakness of the current agent. Tasks that are already solved reliably provide little new training signal. The useful region is therefore tasks that are feasible and realistic, but not yet consistently solved.

   Esakkivel Esakkiraja, Shruthan Radhakrishna, Denis Akhiyarov, Sagar Davasam · Paragraph 14

   AutoSynthData: Generating Training Data for Enterprise Agents · Oct 2, 2026

   Read in source context →

   Continue exploring

   curriculum design →
2. 02

   Feasibility as a core task property

   A useful agentic task must be feasible. At least one trajectory in the current environment must satisfy the user prompt while respecting the system specification. This excludes tasks that require unavailable tools, inaccessible knowledge, impossible state transitions, or policy-prohibited actions.

   Supporting evidence 1

   Original excerpt

   Feasibility. There should exist at least one trajectory in the current environment that satisfies the user prompt while respecting the system specification. This rules out tasks that depend on unavailable tools, inaccessible knowledge, impossible state transitions, or actions prohibited by policy.

   Esakkivel Esakkiraja, Shruthan Radhakrishna, Denis Akhiyarov, Sagar Davasam · Paragraph 12

   AutoSynthData: Generating Training Data for Enterprise Agents · Oct 2, 2026

   Read in source context →

   Continue exploring

   synthetic data quality →
3. 03

   Enterprise agents require environment-specific training

   Enterprises need agents that work well in their own environments—shaped by their systems, rules, and data state. A broadly capable model may still struggle with specific workflows, tool combinations, or constraints unique to that environment.

   Supporting evidence 1

   Original excerpt

   Enterprises need agents that work well in their own environments. The work they ask these agents to do is shaped by the systems they use, the rules they follow, and the state of their data. A model may be broadly capable and still struggle with a particular environment: a workflow it handles poorly, a combination of tools it misuses, or a constraint it fails to respect. Those are the weaknesses an enterprise needs to improve.

   Esakkivel Esakkiraja, Shruthan Radhakrishna, Denis Akhiyarov, Sagar Davasam · Paragraph 1

   AutoSynthData: Generating Training Data for Enterprise Agents · Oct 2, 2026

   Read in source context →

   Continue exploring

   enterprise agent training →